Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Feature Space Particle Inference for Neural Network Ensembles
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| perform particle inference in feature space rather than train independent Deep Ensembles | Wide ResNet-16-4 ensemble · CIFAR-100 | accuracy: 82.3 -> 82.9 | Helped |
| perform particle inference in feature space rather than train independent Deep Ensembles | Wide ResNet-16-4 ensemble · CIFAR-10 | accuracy: 96.4 -> 96.5 | Helped |
| use feature-WGD instead of Deep Ensembles | Wide ResNet-16-4 ensemble · CIFAR-100 | NLL: 0.632 -> 0.624 | Helped |
| use function-space WGD instead of feature-space WGD | Wide ResNet-16-4 ensemble · CIFAR-100 | accuracy: 82.9 -> 79.0 | Hurt |
| use weight-space WGD instead of feature-space WGD | Wide ResNet-16-4 ensemble · CIFAR-100 | accuracy: 82.9 -> 82.3 | Hurt |
| use feature-WGD rather than Deep Ensembles | ResNet-50 ensemble · ImageNet | accuracy: 78.0 -> 78.0 | No clear effect |
| replace Cauchy prior with a uniform prior (ablation) | feature-WGD · CIFAR-100 | accuracy: 82.9 -> 82.1 | Hurt |
| replace Cauchy prior with a normal prior (ablation) | feature-WGD · CIFAR-100 | accuracy: 82.9 -> 82.7 | Hurt |
| remove the feature prior and ensemble interactions, using a single feature-MAP model (ablation) | feature-WGD · CIFAR-100 | accuracy: 82.9 -> 77.6 | Hurt |
| set projection dimension r to 3 instead of 5 (ablation) | feature-WGD · CIFAR-100 | accuracy: 82.9 -> 82.7 | Hurt |
| set projection dimension r to 10 instead of 5 (ablation) | feature-WGD · CIFAR-100 | accuracy: 82.9 -> 82.7 | Hurt |
| remove projection from the repulsive term (ablation) | feature-WGD · CIFAR-100 | accuracy: 82.9 -> 82.7 | Hurt |
Closest papers by meaning
- Input-gradient space particle inference for neural network ensembles (ICLR 2024 · similarity 0.91)
- Function Space Particle Optimization for Bayesian Neural Networks (ICLR 2019 · similarity 0.88)
- Towards Inference Efficient Deep Ensemble Learning (AAAI 2023 · similarity 0.85)
- Connecting the Dots: Is Mode-Connectedness the Key to Feasible Sample-Based Inference in Bayesian Neural Networks? (ICML 2024 · similarity 0.84)
- Microcanonical Langevin Ensembles: Advancing the Sampling of Bayesian Neural Networks (ICLR 2025 · similarity 0.84)
- Path-Guided Particle-based Sampling (ICML 2024 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.